The request by Valve for a better GL_ARB_debug_output implementation for Mesa dates back to last July.
Following Intel and Valve collaborating about Linux GPU drivers, Intel made some driver hacks to expose some debugging/performance information hidden behind an environment variable but didn't toy around with the GL_ARB_debug_output extension. Intel's Eric Anholt is now finally getting to properly supporting this OpenGL debug extension.
Per the Khronos specification on GL_ARB_debug_output, here's some of the important parts:
This extension allows the GL to notify applications when various events occur that may be useful during application development and debugging.Anholt shared on Friday evening that he has an initial GL_ARB_debug_output implementation for Mesa. Right now this work consists of fourteen patches against core Mesa.
These events are represented in the form of enumerable messages with a human-readable string representation. Examples of debug events include incorrect use of the GL, warnings of undefined behavior, and performance warnings.
The type of the message roughly identifies the nature of the event that caused the message. Examples include errors, performance warnings, or warnings about undefined behavior.
Each message is also assigned to a severity level that denotes roughly how "important" that message is in comparison to other messages across all sources and types. For example, notification of a GL error would likely have a higher severity than a performance warning due to redundant state changes.
Messages are communicated to the application through an application- defined callback function that is called by the GL implementation on each debug message. The motivation for the callback routine is to free application developers from actively having to query whether a GL error, or any other debuggable event has happened after each call to a GL function. With a callback, developers can keep their code free of debug checks, and only have to react to messages as they occur. In situations where using a callback is not possible, a message log is also provided that stores copies of recent messages until they are actively queried. To control the volume of debug output, messages can be disabled either individually by ID, or entire groups of messages can be turned off based on combination of source and type.
Among the items hooked into reporting for this debug output extension is the Intel WARN_ONCE macro and the perf_debug() function (formerly the fallback_debug() function), which means any of the previously exposed Intel driver performance information is now reported to this new extension. There was also extra checks added for stalling on mapping a buffer object and shader errors and warnings.
With this work coming from Intel's Open-Source Technology Center, obviously it was designed and hooked up primarily for the i965 DRI driver (though the infrastructure and core functionality in Mesa itself) but other Mesa/Gallium3D drivers will be able to build their support atop these patches. The patch series begins here on the mailing list.